Data processing method and device and electronic equipment

Abstract

The invention discloses a data processing method and device and electronic equipment, and the method comprises the steps: obtaining first watching data of a first target video, the first watching data being data generated when a target object watches the first target video, and being related to at least one operation executed by the target object for the watched first target video; inputting the first target video and the first watching data into a first target model to obtain a first output result of the first target model, the first output result at least being capable of representing a corresponding relationship between at least one multi-modal feature and at least one operation in the first target video; determining a target feature from the at least one multi-modal feature based on the first output result, wherein the influence of the target feature on the first viewing data is greater than the influence of other multi-modal features on the first viewing data; and determining a processing result based on the target feature, wherein the processing result is a processing result of the first target data and the first viewing data.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer, and particularly relates to a data processing method and device and electronic equipment. BACKGROUND

[0002] After a video file is created and published, a video creator usually needs to determine whether the content in the video file is popular according to feedback of a user watching the video file, such as a number of likes, comment content, etc., so as to provide a basis for the video creator in a subsequent video file creation process.

[0003] However, at present, the video creator usually needs to query data of the feedback of the user watching the video file by himself / herself, and manually processes and analyzes the video file itself and the data of the feedback of the user watching the video file by himself / herself, and the manual processing and analysis by the creator himself / herself may cause inaccurate or incomplete results. SUMMARY

[0004] Therefore, the present application provides a data processing method, device and electronic equipment, and specific solutions are as follows.

[0005] A data processing method comprises the following steps.

[0006] Obtaining first viewing data of a first target video, the first viewing data being data generated when a target object watches the first target video, the first viewing data being related to at least one operation performed by the target object for the watched first target video;

[0007] Inputting the first target video and the first viewing data into a first target model to obtain a first output result of the first target model, the first output result being capable of representing a corresponding relationship between at least one multi-modal feature in the first target video and the at least one operation;

[0008] Determining a target feature from the at least one multi-modal feature of the first target video based on the first output result, the target feature having a greater influence on the first viewing data than other multi-modal features;

[0009] Determining a processing result based on the target feature, the processing result being a processing result of the first target video and the first viewing data.

[0010] Further, after the first target video and the first viewing data are input into the first target model, the method further comprises the following steps.

[0011] Obtain a second output result of the first target model output, and the second output result can at least represent adjustment information of at least one multi-modal feature in the first target video based on the first viewing data.

[0012] Further, further comprising:

[0013] Adjust parameters of at least one multi-modal feature in the first target video based on the second output result to obtain an adjusted first target video, and determine the adjusted first target video as a second target video;

[0014] Obtain second viewing data of the second target video, and the second viewing data is generated when the target object watches the second target video, and the second viewing data is related to operations performed by the target object for watching the second target video;

[0015] Compare the first viewing data and the second viewing data to obtain a comparison result, so as to update the first target model based on the comparison result.

[0016] Further, the comparison of the first viewing data and the second viewing data to obtain a comparison result, so as to update the first target model based on the comparison result, comprises:

[0017] Compare the first viewing data and the second viewing data to obtain a comparison result;

[0018] If it is determined that the comparison result represents that the preference value of the target object for the second target video is greater than the preference value for the first target video, the first target model is not updated, and the preference value is determined based on the viewing data when the target object watches the target video;

[0019] If it is determined that the comparison result represents that the preference value of the target object for the first target video is greater than the preference value for the second target video, the first target model is updated.

[0020] Further, the inputting the first target video and the first viewing data into the first target model to obtain a first output result of the first target model comprises:

[0021] Input the first target video and the first viewing data into the first target model;

[0022] Determine, by the first target model, a corresponding video segment of each operation in the at least one operation in the first target video, and the video segment is a part of the first target video;

[0023] The multimodal features of the video segment are determined by the first target model, the correspondence between the multimodal features and the at least one operation is determined, and the correspondence is determined as the first output result.

[0024] Furthermore, determining the video segment corresponding to each of the at least one operations in the first target video using the first target model includes at least one of the following:

[0025] The video segment corresponding to the target exit time period in the first target video is determined by the first target model, wherein the target exit time period is: the time period during which the target object performs a video switching operation in the first target video;

[0026] The video segment corresponding to the target interaction period in the first target video is determined by the first target model, wherein the target interaction video is the period in the first target video during which the target object performs an interactive operation.

[0027] Furthermore, determining the video segment corresponding to the target jump-out time period in the first target video using the first target model includes:

[0028] The frequency of video switching operations performed during different time periods of the first target video is determined by the first target model;

[0029] Based on a time series analysis algorithm, target exit periods are determined in different time periods of the first target video. The frequency of video switching operations performed during the target exit periods is greater than the frequency of video switching operations performed in other time periods besides the target exit periods.

[0030] Furthermore, determining the video segment corresponding to the target interaction period in the first target video using the first target model includes:

[0031] The frequency of interactive operations performed by the target object during different time periods of the first target video is determined by the first target model.

[0032] The target algorithm determines the target interaction period in the first target video, wherein the frequency of interactive operations performed during the target interaction period is greater than the frequency of interactive operations performed during other interaction periods in different time periods excluding the target interaction period. The target algorithm is a clustering algorithm or an anomaly detection algorithm.

[0033] A data processing apparatus, comprising:

[0034] The first obtaining unit is configured to obtain first viewing data of a first target video, the first viewing data being data generated when a target object views the first target video, the first viewing data being related to at least one operation performed by the target object on the viewed first target video;

[0035] The second obtaining unit is configured to input the first target video and the first viewing data into a first target model, and obtain a first output result of the first target model, the first output result being capable of at least representing a corresponding relationship between at least one multi-modal feature in the first target video and the at least one operation;

[0036] The first determining unit is configured to determine a target feature from the at least one multi-modal feature in the first target video based on the first output result, the target feature having a greater impact on the first viewing data than other multi-modal features;

[0037] The second determining unit is configured to determine a processing result based on the target feature, the processing result being a processing result of the first target video and the first viewing data.

[0038] An electronic device includes:

[0039] The processor is configured to obtain first viewing data of a first target video, the first viewing data being data generated when a target object views the first target video, the first viewing data being related to at least one operation performed by the target object on the viewed first target video; input the first target video and the first viewing data into a first target model, and obtain a first output result of the first target model, the first output result being capable of at least representing a corresponding relationship between at least one multi-modal feature in the first target video and the at least one operation; determine a target feature from the at least one multi-modal feature in the first target video based on the first output result, the target feature having a greater impact on the first viewing data than other multi-modal features; and determine a processing result based on the target feature, the processing result being a processing result of the first target video and the first viewing data.

[0040] The memory is configured to store programs required by the processor to perform the above processing process.

[0041] As can be seen from the above technical solutions, the data processing method, apparatus, and electronic device disclosed in this application obtain first viewing data of a first target video. The first viewing data is data generated when a target object watches the first target video. The first viewing data is related to at least one operation, which is performed by the target object on the first target video being watched. The first target video and the first viewing data are input into a first target model to obtain a first output result of the first target model. The first output result can at least characterize the correspondence between at least one multimodal feature of the first target video and at least one operation. Based on the first output result, a target feature is determined from at least one multimodal feature. The target feature has a greater impact on the first viewing data than other multimodal features. Based on the target feature, a processing result is determined, which is the processing result of the first target video and the first viewing data. This solution generates first viewing data by obtaining at least one operation performed by a target object on the first target video. It then uses a first target model to determine the correspondence between at least one multimodal feature in the first target video and the operation performed by the target object. Based on this correspondence, target features are determined, thereby identifying the features in the first target video that affect the first viewing data. This yields the processing results of the first target video and the first viewing data, providing guidance for the creator of the first target video in producing subsequent videos. Furthermore, this solution uses the first target model to automatically analyze and process the first target video and the first viewing data, ensuring the comprehensiveness and accuracy of the data analysis and improving analysis efficiency. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart of a data processing method disclosed in an embodiment of this application;

[0044] Figure 2 This is a flowchart of a data processing method disclosed in an embodiment of this application;

[0045] Figure 3 This is a flowchart of a data processing method disclosed in an embodiment of this application;

[0046] Figure 4 This is a schematic diagram of the structure of a data processing device disclosed in an embodiment of this application;

[0047] Figure 5 This is a schematic diagram of the structure of an electronic device disclosed in an embodiment of this application. Detailed Implementation

[0048] The embodiments of this application are described below with reference to the accompanying drawings. The terminology used in the implementation section of this application is for explaining specific embodiments only and is not intended to limit the scope of this application.

[0049] The embodiments of this application will now be described with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are equally applicable to similar technical problems.

[0050] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0051] This application discloses a data processing method, the flowchart of which is as follows: Figure 1 As shown, it includes:

[0052] Step S11: Obtain the first viewing data of the first target video. The first viewing data is the data generated when the target object watches the first target video. The first viewing data is related to at least one operation, which is performed by the target object for the first target video being watched.

[0053] Step S12: Input the first target video and the first viewing data into the first target model to obtain the first output result of the first target model. The first output result can at least characterize the correspondence between at least one multimodal feature of the first target video and at least one operation.

[0054] Step S13: Based on the first output result, determine the target feature from at least one multimodal feature of the first target video, wherein the target feature has a greater impact on the first viewing data than other multimodal features have an impact on the first viewing data;

[0055] Step S14: Determine the processing result based on the target features. The processing result is the processing result of the first target video and the first viewing data.

[0056] After a video creator completes and publishes a video file, they usually need to rely on feedback from users who watch the video to determine whether the content is popular, such as the number of likes and comments, so as to provide a basis for the creator's subsequent video creation process.

[0057] However, currently, video creators usually need to query user feedback data when watching video files themselves, and manually process and analyze the video files themselves and the user feedback data. This manual processing and analysis by the creators may lead to inaccurate or incomplete results.

[0058] Based on this, in this solution, after the first target video is published, first viewing data generated when the target object watches the first target video is collected. The first viewing data is related to at least one operation, which is performed by the target object in response to the first target video being watched. The at least one operation may include video switching operations and interactive operations. The first target video and the first viewing data are analyzed and processed through a first target model to determine the correspondence between at least one multimodal feature in the first target video and at least one operation, i.e., the first output result. Based on this, target features are determined so as to provide guidance for the production of subsequent videos using the target features. Furthermore, the first target model realizes the automatic processing and analysis of the first viewing data corresponding to the first target video without manual analysis, while also ensuring the comprehensiveness and accuracy of the analysis.

[0059] Once the first target video is produced and published, it can be seen by multiple viewers (target audiences). For example, a short video published on a short video platform (such as Douyin, Kuaishou, etc.) can be seen by users of that short video platform (target audiences), or a long video published on a long video platform (such as iQiyi, Tencent Video, etc.) can be seen by users of that long video platform (target audiences).

[0060] Whether it's a short or long video, after a user (the target audience) sees the video, the user will perform corresponding operations on the video, such as: video switching operations, that is, no longer watching the current video and closing the video interface, or switching to the next video; or interactive operations, such as liking, commenting, sending bullet comments, playing at double speed, skipping some frames of the video, etc. In addition, it should be noted that at least one operation in this embodiment is performed by the target audience on the first target video being watched. Therefore, at least one operation does not include the action of watching the target video.

[0061] Therefore, it is necessary to obtain the first viewing data of the first target video. The first viewing data is generated based on at least one operation performed by the target object on the first target video being viewed. In addition, the first viewing data can be the viewing data of one target object or the viewing data of multiple target objects.

[0062] When obtaining the first viewing data, it is necessary to obtain not only at least one operation, such as video switching and interactive operations, but also the time corresponding to each operation, such as the time corresponding to the video switching operation, the time corresponding to the interactive operation, that is, the time when the target object performs the like operation in the first target video, the time when the target object performs the comment or bullet screen operation in the first target video, and the specific content of the post; in addition, the obtained first viewing data may also include viewing duration data, such as: complete viewing, exiting midway, and the specific time of exiting midway, etc.

[0063] The actions a user takes on a video typically depend on their level of interest in it. If they are not interested, they may directly switch to the next video. If they are interested, they may like, comment, or leave a bullet screen message. Alternatively, they may switch to the next video after it has finished playing. Therefore, after a video is published on the platform, it is necessary to collect statistics on the actions taken by users who watch the video. This means obtaining the first viewing data of the first target video after it is published. The first viewing data is data generated when one or more target objects watch the first target video, and it is related to at least one action taken by the target objects when watching the first target video. The actions include at least video switching and interactive actions.

[0064] After a user watches a video, the action they take depends on their level of interest in the video. The multimodal features of the video itself can influence the user's level of interest, which is reflected in the actions the user takes when watching the video. In other words, the multimodal features of the video itself can influence the actions the user takes when watching the video. Therefore, the correspondence between at least one multimodal feature and at least one action in the first target video can be determined based on the first target video and the first viewing data of the first target video.

[0065] It should be noted that the multimodal features involved in this embodiment may include at least one of the following: editing features, narration features, and background music features. That is, the editing features of a video will affect the user's level of interest in the video, the narration features of a video will also affect the user's level of interest in the video, and the background music features of a video will also affect the user's level of interest in the video.

[0066] In this embodiment, the correspondence between at least one feature and at least one operation in the first target video can be determined by the first target model. Of course, it can also be achieved by other means, such as software or programs, which are not specifically limited here.

[0067] When determining the correspondence between multimodal features in a video and user actions on the video through a first target model, the first target model can specifically be an AI large model, such as the GPT large model (Generative Pre-trained Transformer). The GPT large model is an AI model that can understand and generate natural language by pre-training on massive amounts of text data.

[0068] The first target video and its first viewing data are input into the first target model to obtain the output of the first target model, i.e., the first output result. The first output result represents the correspondence between at least one multimodal feature included in the first target video and at least one operation performed by the target object on the first target video. For example, if the target object performs a video switching operation at the 23rd second of the first target video, that is, at the 23rd second, the target object determines that it is not interested in the first target video, then it is necessary to determine which multimodal features exist in the time period of the 23rd second of the first target video, and which features may cause the target object to perform a video switching operation, such as: during this time period, the statements promoted in the first target video may not be accepted by the target object, etc.

[0069] In addition, the first target model can determine the relationship between one or more operations that can be performed on a certain target object and the multimodal features in the first target video. It can also be an analysis of all operations performed on the first target video by all target objects that see the first target video, so as to ensure the accuracy of the first output result of the first target model.

[0070] After obtaining the first output of the first target model, the target feature (which can be an editing feature, narration feature, or background music feature) is determined from at least one multimodal feature of the first target video using the first output. This target feature provides a basis for subsequent video production. Each multimodal feature of the first target video will affect the first viewing data. Among these, the target feature selected from at least one multimodal feature usually has a greater impact on the first viewing data than other multimodal features. The impact of multimodal features on the first viewing data can include influencing the number of interactions, the viewing duration, etc.

[0071] For example, taking the impact of background music features on the number of likes, when the background music feature corresponds to the first background music in the first target video, its number of likes is 20; when the background music feature of the first target video corresponds to the second background music, its number of likes is 2000. Regarding the impact of commentary features on the number of likes, when the commentary feature corresponds to the first commentary in the first target video, its number of likes is 20; when the commentary feature of the first target video corresponds to the second commentary, its number of likes is 200. Therefore, the impact of background music features on the number of likes for the first target video is greater than the impact of commentary features.

[0072] Among them, the target features can be: the multimodal features of the first target model that cause the number of likes of the first target video to be greater than a certain threshold, or the multimodal features of the first target model that cause the bounce rate of the first target model to be less than a certain threshold in a certain period of time (the bounce rate is the frequency at which the target object performs video switching operations in that period of time), etc.

[0073] After determining the target characteristics, the processing results for the first target video and the first viewing data can be determined based on the target characteristics. Based on the processing results, the creator of the first target video can provide guidance for the production of subsequent videos, or adjust the first target video based on the processing results to achieve the purpose of adjusting the target audience's first viewing data of the first target video, such as increasing the number of views, increasing the number of likes, increasing the viewing time, etc.

[0074] The data processing method disclosed in this embodiment obtains first viewing data generated by at least one operation performed by a target object on the first target video when the target object watches the first target video. It then uses a first target model to determine the correspondence between at least one multimodal feature in the first target video and the operation performed by the target object. Based on this correspondence, target features are determined, thereby identifying the features in the first target video that affect the first viewing data. This yields the processing results of the first target video and the first viewing data, which can then be used to guide the creator of the first target video in the production of subsequent videos. Furthermore, this solution uses a first target model to achieve automatic analysis and processing of the first target video and the first viewing data, ensuring the comprehensiveness and accuracy of the data analysis and processing, and improving analysis efficiency.

[0075] This embodiment discloses a video processing method, the flowchart of which is as follows: Figure 2 As shown, it includes:

[0076] Step S21: Obtain first viewing data of the first target video. The first viewing data is the data generated when the target object watches the first target video. The first viewing data is related to at least one operation, which is performed by the target object for the first target video being watched.

[0077] Step S22: Input the first target video and the first viewing data into the first target model;

[0078] Step S23: Obtain the first output result of the first target model. The first output result can at least characterize the correspondence between at least one multimodal feature and at least one operation in the first target video.

[0079] Step S24: Based on the first output result, determine the target feature from at least one multimodal feature of the first target video, where the target feature has a greater impact on the first viewing data than other multimodal features.

[0080] Step S25: Determine the processing result based on the target features. The processing result is the processing result of the first target video and the first viewing data.

[0081] Step S26: Obtain the second output result of the first target model. The second output result can at least characterize the adjustment information of at least one multimodal feature in the first target video.

[0082] After the first target video is published and can be seen by other users, the data generated when the target object watches the first target video is statistically analyzed to obtain the first viewing data. The generation of the first viewing data is related to at least one operation performed by the target object on the first target video. Then, the first target video and the first viewing data are input into the first target model. The first target model can output a first output result. The first output result can characterize the correspondence between at least one multimodal feature in the first target video and the operation performed by the target object on the first target video. The target features can be determined through the first output result, so as to provide a reference for subsequent video generation using the target features.

[0083] In the data processing method disclosed in this embodiment, after inputting the first target video and the first viewing data into the first target model, not only can the first output result output by the first target model be obtained, but also the second output result output by the first target model can be obtained. The second output result can at least characterize the adjustment information of at least one multimodal feature in the first target video.

[0084] Specifically, after the first target video and the first viewing data are input into the first target model, the first target model analyzes and processes the first target video and the first viewing data to determine which multimodal features in the first target video will affect the operations performed by the target object. For example, which multimodal features will lead to an increase in the target object's video switching operations, which multimodal features will lead to an increase in the target object's like operations, and which multimodal features will lead to a decrease in the target object's comment operations, etc. This information is statistically analyzed to obtain the first output result, which represents the correspondence between different multimodal features and operations in the first target video.

[0085] The first target model obtains the correspondence between different multimodal features and operations in the first target video, and outputs a first output result so that the data processing device on which the data processing method disclosed in this embodiment is based can determine the target feature from multiple multimodal features in the first target video based on the first output result. After that, the first target model continues to analyze and process based on the first output result in order to obtain adjustment information for different multimodal features in the first target video.

[0086] The first output represents the correspondence between different multimodal features and operations in the first target video. Therefore, to change the operations performed by the target audience on the first target video, the multimodal features in the first target video need to be adjusted. For example, if a first feature in the first target video leads to an increase in video switching operations performed by the target audience, then to avoid losing viewers, the first feature needs to be adjusted. This could be done by directly deleting the first feature, or by modifying it to a second feature to reduce the video switching operations performed by the target audience due to the first feature, thereby increasing the viewing time of the first target video or increasing the target audience's interest in the first target video.

[0087] Therefore, while obtaining and outputting the first output result, the first target model can continue to perform analysis and processing based on the first output result in order to determine the adjustment information for at least one multimodal feature in the first target video.

[0088] In addition, the data processing method disclosed in this embodiment may also be as follows: after the first target model outputs a first output result, the target features are determined based on the first output result, and then the target features are input into the first target model so that the first target model determines a second output result based on the target features and the first output result, and outputs the second output result.

[0089] The target feature is the feature that influences the degree of interest of the target object watching the first target video in the first target video, that is, the feature that influences the operation performed by the target object watching the first target video. Furthermore, the target feature is the multimodal feature with the highest influence parameter selected from multiple multimodal features of the first target video based on the first output result. For example, the influence degree of the first feature on the first target video is a first value, the influence degree of the second feature on the first target video is a second value, and the influence degree of the third feature on the first target video is a third value, where the first value < the second value < the third value. Therefore, the third feature corresponding to the third value can be determined as the target feature. Specifically, the influence degree of the first feature on the first target video is as follows: if the first feature is located in the first time period of the first target video, the probability of the target object performing a video switching operation in the first time period is 5%; if the second feature is located in the second time period of the first target video, the probability of the target object performing a video switching operation in the second time period is 10%; and if the third feature is located in the third time period of the first target video, the probability of the target object performing a video switching operation in the third time period is 50%, where 5% < 10% < 50%. Therefore, the third feature can be determined as the feature with the highest influence degree on the video switching operation and is determined as the target feature.

[0090] Since the target feature may be a multimodal feature that increases the video switching operation performed by the target object, or it may be a multimodal feature that decreases the video switching operation performed by the target object, the first target model, after obtaining the target feature, also needs to determine the type of the target feature based on the correspondence between the target feature and at least one operation in the first output result. That is, it needs to determine whether the target feature corresponds to a feature that increases or decreases the video switching operation, a feature that increases or decreases the like operation, etc. If the target feature is a feature that increases the like operation, the second output result can be a suggestion to represent increasing the target feature; if the target feature is a feature that decreases the video switching operation, the second output result is a feature that represents decreasing the target feature, etc. In other words, the first target model determines the second output result based on the type of the target feature.

[0091] For example, the second output could be: "It is recommended to add a narration at the X-second mark to explain the background information of the current scene in order to reduce the confusion of the audience." Another possible second output could be: "It is recommended to speed up the editing pace from the X-second mark to the Y-second mark in order to increase the audience's viewing interest."

[0092] Alternatively, the data processing method disclosed in this embodiment may also be as follows: after the first target model outputs a first output result, the target features are determined based on the first output result. Then, the target features, the first target video, and the first viewing data are input into the second target model so that the second target model determines the second output result based on the target features and the first output result, and outputs the second output result.

[0093] The data processing method disclosed in this embodiment obtains first viewing data generated based on operations performed on the first target video when a target object watches it. The first target video and the first viewing data are input into a first target model. The first target model can obtain a first output result representing the correspondence between at least one multimodal feature in the first target video and at least one operation. It can also obtain a second output result representing the adjustment information of at least one multimodal feature in the first target video based on the first viewing data. The first output result can be used to determine target features, thereby providing a reference for subsequent video production. The second output result can also be used to adjust at least one multimodal feature in the first target video to improve the target object's viewing experience of the adjusted first target video.

[0094] The data processing method disclosed in this embodiment may further include:

[0095] The parameters of at least one multimodal feature in the first target video are adjusted based on the second output result to obtain the adjusted first target video, which is then identified as the second target video. Second viewing data of the second target video is obtained, which is data generated when the target object watches the second target video and is related to the operation performed by the target object on the watched second target video. The first viewing data and the second viewing data are compared to obtain a comparison result, which is used to update the first target model based on the comparison result.

[0096] The second output represents the adjustment information for at least one multimodal feature in the first target video. After the first target model outputs the second output, the first target video can be adjusted based on the second output to obtain the adjusted first target video, which is then identified as the second target video. After obtaining the second target video, it is published. If the first target video is published on the first platform, then the second target video can be published on the second platform. The first platform and the second platform can be the same platform or different platforms.

[0097] It should be noted that after obtaining the second output result, the second output result can be filtered and optimized, that is, unreasonable or duplicate suggestions in the second output result can be removed to ensure the practicality and operability of the suggestions.

[0098] After the second target video is published, users of the second platform can watch the second target video. At this time, second viewing data of the second target video can be collected, that is, data generated by one or more target objects watching the second target video. The second viewing data can be specifically determined based on the operations performed by one or more target objects on the second target video when watching it.

[0099] After obtaining the second viewing data, the first viewing data and the second viewing data can be compared to obtain a comparison result. This comparison result can characterize the difference between the operation performed by the target object in the first target video before adjustment and the operation performed by the target object in the second target video after adjustment, and determine whether the second output result is reasonable based on this.

[0100] For example: In the first viewing data, the frequency of video switching operation in the first time period of the first target video is a1, and in the second viewing data, the frequency of video switching operation in the first time period of the second target video is b1. Comparing the magnitudes of a1 and b1, if a1 < b1, and the second output result at least represents the adjustment of features in the first time period of the first target video, since the frequency of video switching operation in the first time period increases after the feature adjustment, then the second output result is unreasonable.

[0101] For example, if the frequency of performing a "like" operation on the first target video in the second time period of the first viewing data is a2, and the frequency of performing a "like" operation on the second target video in the second viewing data is b2, and if a2 < b2, and the second output result at least represents the adjustment of the features in the second time period of the first target video, then the second output result is reasonable because the frequency of performing a "like" operation in the second time period increases after the features in the second time period are adjusted.

[0102] After determining whether the second output result is reasonable based on the comparison results of the first viewing data and the second viewing data, the first target model can be updated accordingly to increase the probability that the updated first target model will output a reasonable second output result.

[0103] That is: compare the first viewing data with the second viewing data to obtain the comparison result, so as to update the first target model based on the comparison result, which can be specifically as follows:

[0104] Compare the first viewing data with the second viewing data to obtain the comparison result. If the comparison result indicates that the target object's preference value for the second target video is greater than its preference value for the first target video, the first target model is not updated. The preference value is determined based on the viewing data when the target object watches the target video. If the comparison result indicates that the target object's preference value for the first target video is greater than its preference value for the second target video, the first target model is updated.

[0105] The preference value is determined based on viewing data of the target audience when watching the target video. For example, an increase in the viewing time of the target video will increase the preference value, while a decrease in the viewing time will decrease it. Liking the target video or an increase in the number of likes will increase the preference value, while not liking the target video or a decrease in the number of likes will decrease it. Commenting on the target video with a positive comment will increase the preference value, while not commenting will decrease it. Alternatively, commenting on the target video with a negative comment will decrease the preference value.

[0106] If, based on the comparison between the first and second viewing data, it is determined that the target audience's preference value for the first target video is greater than their preference value for the second target video (e.g., the number of likes for the first target video is greater than the number of likes for the second target video), then the second output result is determined to be unreasonable, and the first target model needs to be updated. If, based on the comparison between the first and second viewing data, it is determined that the target audience's preference value for the second target video is greater than their preference value for the first target video (e.g., the number of positive comments in the second target video is greater than the number of positive comments in the first target video), then the second output result is determined to be reasonable, and the target model does not need to be updated.

[0107] Alternatively, if the second output includes adjustment information for multimodal features in different time periods, then it is necessary to analyze whether the adjustment information for multimodal features in each time period is reasonable in order to update the first target model based on the analysis results; or if the second output includes adjustments for different multimodal features, then it is necessary to analyze whether the adjustments for different multimodal features are reasonable in order to update the first target model based on the analysis results.

[0108] It should be noted that in the data processing method disclosed in this embodiment, the first viewing data and the second viewing data can be compared using the first target model to obtain a comparison result, that is:

[0109] The first viewing data, the second viewing data, and the second output result are input into the first target model to obtain the comparison result output by the first target model.

[0110] The data processing method disclosed in this embodiment adjusts the first target model based on the comparison results of the first viewing data and the second viewing data, so as to improve the performance of the first target model and ensure the accuracy and practicality of the second output result output by the first target model. By continuously optimizing the first target model, it can more accurately generate the second output result that meets the needs of creators and the preferences of the audience.

[0111] This embodiment discloses a data processing method, the flowchart of which is as follows: Figure 3 As shown, it includes:

[0112] Step S31: Obtain the first viewing data of the first target video. The first viewing data is the data generated when the target object watches the first target video. The first viewing data is related to at least one operation, which is performed by the target object for the first target video being watched.

[0113] Step S32: Input the first target video and the first viewing data into the first target model;

[0114] Step S33: Determine the video segment corresponding to each operation in the first target video through the first target model. The video segment is a part of the first target video.

[0115] Step S34: Determine the multimodal features of the video segment through the first target model, determine the correspondence between the multimodal features and at least one operation, and determine the correspondence as the first output result;

[0116] Step S35: Based on the first output result, determine the target feature from at least one multimodal feature of the first target video, wherein the target feature has a greater impact on the first viewing data than other multimodal features have an impact on the first viewing data;

[0117] Step S36: Determine the processing result based on the target features. The processing result is the processing result of the first target video and the first viewing data.

[0118] After the first target video is published and can be seen by other users, the data generated when one or more target objects watch the first target video is statistically analyzed to obtain first viewing data. The generation of the first viewing data is related to at least one operation performed by the target object on the first target video. Then, the first target video and the first viewing data are input into the first target model. The first target model can output a first output result. The first output result can characterize the correspondence between at least one multimodal feature in the first target video and the operation performed by the target object on the first target video. The target features can be determined through the first output result, so as to provide a reference for subsequent video generation using the target features.

[0119] Specifically, the first target model outputs a first output result based on the first target video and the first viewing data. Specifically, the first target model first determines the corresponding video segment in the first target video for each operation performed by the target object based on the first target video and the first viewing data. Then, it determines the multimodal features of the video segment and associates the multimodal features of the video segment with the operation, that is, it determines the correspondence between the multimodal features of the video segment and at least one operation. Thus, the first output result can be obtained.

[0120] The operations performed by one or more target objects on the first target video may be located at different times in the first target video. Therefore, after determining the operations performed by one or more target objects respectively, it is necessary to determine the time in the first target video corresponding to each operation, and further determine the corresponding video segment.

[0121] Since the operation can be a video switching operation or an interactive operation, for the video switching operation, it can be specifically as follows: determine the video segment corresponding to the target jump-out time period in the first target video through the first target model, where the target jump-out time period is: the time period in the first target video during which the target object performs the video switching operation.

[0122] Once it is determined that a video switching operation was performed on the first target video, the time period corresponding to that video switching operation in the first target video is determined, and this time period is designated as the target exit time period. The corresponding video segment within the first target video is also determined. If there is only one video switching operation, there is one corresponding target exit time period and one corresponding video segment; if there are multiple video switching operations, there are multiple corresponding target exit time periods and multiple corresponding video segments.

[0123] Alternatively, instead of determining all time periods corresponding to video switching operations as the target exit time period, one time period is selected from multiple time periods corresponding to video switching operations as the target exit time period. During this target exit time period, the frequency of video switching operations is greater than the frequency of video switching operations during other time periods.

[0124] Specifically, this can be achieved by: determining the frequency of video switching operations performed during different time periods of the first target video when the target object is watching the first target video using the first target model; and determining the target exit time periods in different time periods of the first target video based on a time series analysis algorithm, wherein the frequency of video switching operations performed during the target exit time periods is greater than the frequency of video switching operations performed during other time periods in different time periods excluding the target exit time periods.

[0125] Identify all video switching operations performed on the first target video and determine the time period corresponding to each video switching operation. From this, determine the target exit time period. The frequency of video switching operations performed during the target exit time period is greater than the frequency of video switching operations performed during other time periods. For example, if the frequency of video switching operations performed in the first time period of the first target video is 20, the frequency of video switching operations performed in the second time period of the first target video is 10, and the frequency of video switching operations performed in the third time period of the first target video is 70, then the third time period is the target exit time period of the first target video.

[0126] "Jump out" means to exit from the first target video, switch to another video, or close video playback.

[0127] This involves statistically analyzing the bounce rate of the target object at different time periods, which is the proportion of the target object that exits the first target video at different time periods. Algorithms can be used to identify time periods with significantly higher bounce rates than other time periods, mark them as target bounce periods, and determine the video segments in the first target video corresponding to the target bounce periods.

[0128] Specifically, identifying target bounce periods using algorithms can be achieved by employing time series analysis methods, such as moving average, exponential average, autoregressive models, and moving average models. Taking the moving average method as an example, if the total duration of the first target video is 10 minutes, the bounce rate is calculated per minute. A simple mean average (SMA) smoothing sequence with a window size of 3 is used to calculate the moving average for each time period. If the actual bounce rate exceeds the sum of the moving average and a threshold (e.g., twice the standard deviation), then that time period is identified as the target bounce period.

[0129] There can be one or more target bounce periods. When there is only one target bounce period, the period with the highest bounce rate is selected from multiple periods as the target bounce period. When there are multiple target bounce periods, the period with a bounce rate higher than a certain value is selected as the target bounce period.

[0130] For interactive operations, the specific steps are as follows: determine the video segment corresponding to the target interactive time period in the first target video through the first target model, wherein the target interactive video is the time period in which the target object performs interactive operations in the first target fourteen items.

[0131] Once it is determined that an interactive operation was performed on the first target video, the time period corresponding to the interactive operation in the first target video is determined, and this time period is designated as the target interactive time period. The corresponding video segment within the first target video is also determined for this target interactive time period. If there is only one interactive operation, there is one corresponding target interactive time period and one corresponding video segment; if there are multiple interactive operations, there are multiple corresponding target interactive time periods and multiple corresponding video segments.

[0132] Alternatively, instead of determining all time periods corresponding to the execution of interactive operations as the target interactive time period, one time period is selected from the time periods corresponding to multiple execution of interactive operations as the target interactive time period. During the target interactive time period, the frequency of executing interactive operations is greater than the frequency of executing interactive operations during the time periods corresponding to other execution of interactive operations.

[0133] Specifically, this can be achieved by: determining the frequency of interactive operations performed by the target object during different time periods of the first target video when watching the first target video using the first target model; and determining the target interaction time periods in different time periods of the first target video based on the target algorithm, wherein the frequency of interactive operations performed during the target interaction time periods is greater than the frequency of interactive operations performed during other interaction time periods in different time periods, and the target algorithm is a clustering algorithm or an anomaly detection algorithm.

[0134] Identify all interactive operations performed on the first target video and determine the time period corresponding to each interactive operation. From this, determine the target interactive time period. The frequency of performing interactive operations during the target interactive time period is greater than the frequency of performing interactive operations during other time periods. For example, if the frequency of performing interactive operations during the first time period of the first target video is 30, the frequency of performing interactive operations during the second time period of the first target video is 10, and the frequency of performing interactive operations during the third time period of the first target video is 60, then the third time period is the target interactive time period of the first target video.

[0135] Interactive actions may include: liking, posting comments, and posting bullet comments.

[0136] It can count the time in the first target video corresponding to all interactive operations, determine the frequency of interactive operations in different time periods, identify time periods with significantly higher interaction frequency than other time periods through algorithms, mark them as target interaction time periods, and determine the video segments of the first target video corresponding to the target interaction time periods.

[0137] Specifically, identifying target interaction periods through algorithms can be achieved by using clustering algorithms or anomaly detection algorithms.

[0138] It should be noted that there can be one or more target interaction periods. When there is one target interaction period, the period with the highest interaction frequency is selected from multiple periods as the target interaction period. When there are multiple target interaction periods, the period with the interaction frequency higher than a certain value is determined as the target interaction period.

[0139] Alternatively, determining video segments can also be done as follows: after determining the target bounce time and the target interaction time, determine the lost video segments in the first target video based on the target bounce time and the target interaction time. The lost video segments are the video segments corresponding to the time when viewers are lost. For highlight video segments, they can be directly determined based on the target interaction time. The highlight video segments are the video segments corresponding to the time when viewers' attention is attracted and interaction is stimulated.

[0140] After determining the video segment, the first target model needs to extract the multimodal features from the video segment and associate the multimodal features of the video segment with the operation to obtain the first output result.

[0141] The multimodal features of a video clip may include at least one of the following: clip features, text features, and audio features. It is possible to determine the correspondence between the clip features of a video clip and the operations performed within that video clip; to determine the correspondence between the text features of a video clip and the operations performed within that video clip; and to determine the correspondence between the audio features of a video clip and the operations performed within that video clip.

[0142] For video clip editing features, the editing rhythm of the video clips can be analyzed, such as using analysis tools to extract shot switching frequency, scene transition time, etc. Editing features can be correlated with the operations performed within that time period to determine the impact of different editing features on bounce rate and interaction frequency. For example, fast editing may cause viewers to feel fatigued and jump out prematurely, while appropriate rhythm variations may help attract viewers' attention.

[0143] For textual features of video clips, the narration text can be extracted, and natural language processing (NLP) techniques can be used to perform sentiment analysis on the narration text to determine its sentiment tendency (positive, negative, neutral). Additionally, the length of the narration text can be determined. Correlation between the sentiment tendency of the narration text and the actions performed within the video clip can be analyzed to understand the impact of the narration text on viewer emotions and viewing behavior. For example, positive narration text may increase viewer interaction frequency, while negative or lengthy narration text may lead to an increased viewer bounce rate.

[0144] For audio features of video clips, the audio features of the background music can be extracted. For example, audio processing tools (such as Librosa) can be used to extract the rhythm, pitch, and emotional style of the background music. These audio features can then be correlated with the operations performed within the video clip to determine their impact on viewer emotions and viewing behavior. For instance, upbeat background music may increase viewer interaction frequency, while background music that doesn't match the video content may lead to a higher bounce rate.

[0145] Associating a single feature of a video segment with the operation performed within that video segment yields the first output result, ensuring accurate analysis of each multimodal feature in the video segment. Alternatively, multiple multimodal features can be combined and associated with the operation performed within the video segment to obtain the first output result, ensuring comprehensive analysis.

[0146] Furthermore, the data processing method disclosed in this embodiment, when obtaining the first viewing data of the first target video, not only needs to obtain the operation performed by the target object on the first target video being viewed, but also needs to obtain the time in the first target video corresponding to the execution of each operation.

[0147] In addition, when obtaining the first viewing data, it is necessary to further preprocess the first viewing data. Preprocessing may include: identifying invalid data in the first viewing data and deleting invalid data. Invalid data may include: duplicate bullet comments, malicious comments, etc.

[0148] Preprocessing may also include: determining the target format, converting the first viewing data to the target format, etc.

[0149] The collected data is converted into a unified format, i.e., the target format, to facilitate subsequent analysis and processing. For example, all moments are converted into a unified time unit, such as seconds, and the content of bullet comments and comments is segmented into words.

[0150] In addition, preprocessing may include data labeling, where some data can be labeled according to actual needs, such as labeling the sentiment (positive, negative, neutral) of bullet comments and comments, to provide a basis for subsequent sentiment analysis and correlation.

[0151] The data processing method disclosed in this embodiment obtains first viewing data generated based on the operations performed on the first target video when the target object watches the first target video. The first target video and the first viewing data are input into a first target model. The first target model is used to determine the video segments in the first target video corresponding to each operation. Furthermore, the first target model is used to associate the multimodal features of the video segments with the operations, thereby obtaining a first output result. This solution achieves layer-by-layer analysis and processing of the first viewing data corresponding to the first target video through the first target model, ensuring the comprehensiveness and accuracy of the analysis and processing, and improving the analysis efficiency.

[0152] This embodiment discloses a data processing device, the structural schematic diagram of which is shown below. Figure 4 As shown, it includes:

[0153] The system comprises a first obtaining unit 41, a second obtaining unit 42, a first determining unit 43, and a second determining unit 44.

[0154] The first obtaining unit 41 is used to obtain first viewing data of the first target video. The first viewing data is data generated when the target object watches the first target video. The first viewing data is related to at least one operation, which is performed by the target object for the first target video being watched.

[0155] The second obtaining unit 42 is used to input the first target video and the first viewing data into the first target model to obtain the first output result of the first target model. The first output result can at least characterize the correspondence between at least one multimodal feature in the first target video and at least one operation.

[0156] The first determining unit 43 is used to determine target features from at least one multimodal feature of the first target video based on the first output result, wherein the target features have a greater impact on the first viewing data than other multimodal features have an impact on the first viewing data;

[0157] The second determining unit 44 is used to determine the processing result based on the target features. The processing result is the processing result of the first target video and the first viewing data.

[0158] Furthermore, the second obtaining unit is also used for:

[0159] Obtain a second output result from the output of the first target model. The second output result can at least characterize the adjustment information of at least one multimodal feature in the first target video based on the first viewing data.

[0160] Furthermore, the data processing apparatus disclosed in this embodiment may further include:

[0161] An update unit is used to determine the adjustment of parameters of at least one multimodal feature in the first target video based on the second output result to obtain the adjusted first target video, and to determine the adjusted first target video as the second target video; to obtain second viewing data of the second target video, wherein the second viewing data is data generated when the target object watches the second target video, and the second viewing data is related to the operation of the target object on the watched second target video; and to compare the first viewing data and the second viewing data to obtain a comparison result so as to update the first target model based on the comparison result.

[0162] Furthermore, the update unit is used for:

[0163] Compare the first viewing data with the second viewing data to obtain a comparison result; if the comparison result indicates that the target object's preference value for the second target video is greater than its preference value for the first target video, do not update the first target model, the preference value is determined based on the viewing data when the target object watches the target video; if the comparison result indicates that the target object's preference value for the first target video is greater than its preference value for the second target video, update the first target model.

[0164] Furthermore, the second obtaining unit is used for:

[0165] The first target video and the first viewing data are input into the first target model; the first target model is used to determine the video segment corresponding to each operation in the first target video, and the video segment is a part of the first target video; the first target model is used to determine the multimodal features of the video segment, and the correspondence between the multimodal features and at least one operation is determined, and the correspondence is determined as the first output result.

[0166] Furthermore, the second obtaining unit is used for:

[0167] The first target model determines the video segment corresponding to the target exit time period in the first target video, where the target exit time period is the time period during which the target object performs a video switching operation in the first target video; the first target model also determines the video segment corresponding to the target interaction time period in the first target video, where the target interaction video is the time period during which the target object performs an interactive operation in the first target video.

[0168] Furthermore, the second obtaining unit is used for:

[0169] The first target model determines the frequency of video switching operations performed by the target object during different time periods of the first target video. Based on a time series analysis algorithm, the target exit time periods within different time periods of the first target video are determined. The frequency of video switching operations performed during these target exit time periods is greater than the frequency of video switching operations performed during other time periods outside of these target exit time periods.

[0170] Furthermore, the second obtaining unit is used for:

[0171] The first target model determines the frequency of interactive operations performed by the target object during different time periods of the first target video when watching the first target video; the target algorithm determines the target interactive time period in the first target video, wherein the frequency of interactive operations performed during the target interactive time period is greater than the frequency of interactive operations performed during other interactive time periods in different time periods excluding the target interactive time period. The target algorithm is a clustering algorithm or an anomaly detection algorithm.

[0172] The data processing apparatus disclosed in this embodiment is implemented based on the data processing method disclosed in the above embodiments, and will not be described again here.

[0173] The data processing apparatus disclosed in this embodiment obtains first viewing data generated by the operation performed by the target object on the first target video when the target object watches the first target video, and uses a first target model to determine the correspondence between at least one multimodal feature in the first target video and the operation performed by the target object. Based on the correspondence, target features are determined, and processing results are obtained based on the target features. This facilitates the provision of guidance for the production of subsequent videos based on the processing results. Furthermore, in this solution, the first target model is used to realize the automatic analysis and processing of the first viewing data corresponding to the first target video, ensuring the comprehensiveness and accuracy of the analysis and processing, and improving the analysis efficiency.

[0174] This embodiment discloses an electronic device, the structural schematic diagram of which is shown below. Figure 5 As shown, it includes:

[0175] Processor 51 and memory 52.

[0176] The processor 51 is configured to obtain first viewing data of a first target video, which is data generated when a target object watches the first target video. The first viewing data is related to at least one operation, which is performed by the target object on the watched first target video. The processor 51 inputs the first target video and the first viewing data into a first target model to obtain a first output result of the first target model. The first output result can at least characterize the correspondence between at least one multimodal feature in the first target video and at least one operation. Based on the first output result, the processor 51 determines a target feature from at least one multimodal feature of the first target video. The target feature has a greater impact on the first viewing data than other multimodal features. Based on the target feature, the processor 51 determines a processing result, which is the processing result of the first target video and the first viewing data.

[0177] The electronic device disclosed in this embodiment is implemented based on the data processing method disclosed in the above embodiments, and will not be described again here.

[0178] The electronic device disclosed in this embodiment obtains first viewing data generated by the operation performed by the target object on the first target video when the target object watches the first target video, and uses a first target model to determine the correspondence between at least one multimodal feature in the first target video and the operation performed by the target object. Based on the correspondence, target features are determined, and further processing results are obtained, which facilitates the provision of guidance for the production of subsequent videos based on the processing results. Furthermore, in this solution, the first target model is used to realize the automatic analysis and processing of the first viewing data corresponding to the first target video, which ensures the comprehensiveness and accuracy of the analysis and processing and improves the analysis efficiency.

[0179] This application embodiment also provides a readable storage medium on which a computer program is stored. The computer program is loaded and executed by a processor to implement the steps of the above data processing method. The specific implementation process can be referred to the description of the corresponding part of the above embodiment, and will not be repeated in this embodiment.

[0180] This application also proposes a computer program product or computer program that includes computer instructions stored in a computer-readable storage medium. The processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the methods provided in the various optional implementations of the data processing method or data processing apparatus described above. Specific implementation processes can be referred to the descriptions of the corresponding embodiments above, and will not be repeated here.

[0181] It should also be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. In addition, in the device embodiment drawings provided in this application, the connection relationship between modules indicates that they have a communication connection, which can be implemented as one or more communication buses or signal lines.

[0182] Through the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware, or it can be implemented by special-purpose hardware including application-specific integrated circuits, special-purpose CPUs, special-purpose memory, special-purpose components, etc. Generally, any function performed by a computer program can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be diverse, such as analog circuits, digital circuits, or special-purpose circuits. However, for this application, software program implementation is more often the preferred implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a readable storage medium, such as a computer floppy disk, USB flash drive, mobile hard disk, ROM, RAM, magnetic disk, or optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, training equipment, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0183] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product.

[0184] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions may be transmitted from one website, computer, training device, or data center to another website, computer, training device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state drives (SSDs)).

Claims

1. A data processing method, characterized by, The method comprises: obtaining first viewing data of a first target video, the first viewing data being data generated when a target object views the first target video, the first viewing data being related to at least one operation performed by the target object on the viewed first target video; inputting the first target video and the first viewing data into a first target model to obtain a first output result of the first target model, the first output result being capable of representing at least a correspondence between at least one multi-modal feature of the first target video and the at least one operation; determining a target feature from the at least one multi-modal feature of the first target video based on the first output result, the target feature having a greater impact on the first viewing data than other multi-modal features; determining a processing result based on the target feature, the processing result being a processing result of the first target video and the first viewing data.

2. The method of claim 1, wherein, After inputting the first target video and the first viewing data into the first target model, the method further comprises: obtaining a second output result output by the first target model, the second output result being capable of representing at least adjustment information for at least one multi-modal feature in the first target video.

3. The method of claim 2, wherein, The method further comprises: adjusting a parameter of the at least one multi-modal feature in the first target video based on the second output result to obtain an adjusted first target video, and determining the adjusted first target video as a second target video; obtaining second viewing data of the second target video, the second viewing data being data generated when the target object views the second target video, the second viewing data being related to an operation performed by the target object on the viewed second target video; comparing the first viewing data and the second viewing data to obtain a comparison result, so as to update the first target model based on the comparison result.

4. The method of claim 3, wherein, The comparing the first viewing data and the second viewing data to obtain a comparison result, so as to update the first target model based on the comparison result, comprises: comparing the first viewing data and the second viewing data to obtain a comparison result; if it is determined that the comparison result represents that a preference value of the target object for the second target video is greater than a preference value for the first target video, not updating the first target model, the preference value being determined based on viewing data when the target object views a target video; if it is determined that the comparison result represents that a preference value of the target object for the first target video is greater than a preference value for the second target video, updating the first target model.

5. The method of claim 1, wherein, The inputting the first target video and the first viewing data into the first target model to obtain a first output result of the first target model comprises: inputting the first target video and the first viewing data into the first target model; determining, by the first target model, a video segment corresponding to each operation in the at least one operation in the first target video, the video segment being a part of the first target video; determine a multi-modal feature of the video clip through the first target model, determine a corresponding relationship between the multi-modal feature and the at least one operation, and determine the corresponding relationship as a first output result.

6. The method of claim 5, wherein, The determination of the video clip corresponding to each operation in the at least one operation in the first target video through the first target model at least includes one of the following: determine a video clip corresponding to a target jump-out period in the first target video through the first target model, wherein the target jump-out period is a period in which a target object performs a video switching operation in the first target video. determine a video clip corresponding to a target interaction period in the first target video through the first target model, wherein the target interaction period is a period in which the target object performs an interaction operation in the first target video.

7. The method of claim 6, wherein, The determination of the video clip corresponding to the target jump-out period in the first target video through the first target model includes: determine a frequency of performing a video switching operation in different periods of the first target video when the target object watches the first target video through the first target model; determine a target jump-out period in different periods of the first target video based on a time series analysis algorithm, wherein the frequency of performing the video switching operation in the target jump-out period is greater than the frequency of performing the video switching operation in other periods of the different periods except the target jump-out period.

8. The method of claim 6, wherein the determination of the video clip corresponding to the target interaction period in the first target video through the first target model includes: determine a frequency of performing an interaction operation in different periods of the first target video when the target object watches the first target video through the first target model; determine a target interaction period in the first target video through a target algorithm, wherein the frequency of performing the interaction operation in the target interaction period is greater than the frequency of performing the interaction operation in other interaction periods of the different periods except the target interaction period, and the target algorithm is a clustering algorithm or an anomaly detection algorithm.

9. A data processing apparatus, characterized by, The method includes: a first obtaining unit configured to obtain first watching data of a first target video, the first watching data being data generated when a target object watches the first target video, the first watching data being related to at least one operation performed by the target object for the first target video; a second obtaining unit configured to input the first target video and the first watching data to a first target model, and obtain a first output result of the first target model, the first output result being capable of representing a corresponding relationship between at least one multi-modal feature in the first target video and the at least one operation; a first determining unit configured to determine a target feature from the at least one multi-modal feature in the first target video based on the first output result, the target feature having a greater influence on the first watching data than other multi-modal features. A second determining unit is configured to determine a processing result based on the target feature, the processing result being a processing result of the first target video and the first viewing data.

10. An electronic device, comprising: The method comprises the following steps: A processor is configured to obtain first viewing data of a first target video, the first viewing data being data generated when a target object watches the first target video, the first viewing data being related to at least one operation performed by the target object for watching the first target video; input the first target video and the first viewing data into a first target model to obtain a first output result of the first target model, the first output result being capable of representing a corresponding relationship between at least one multi-modal feature in the first target video and the at least one operation; determine a target feature from the at least one multi-modal feature of the first target video based on the first output result, the target feature having a greater impact on the first viewing data than other multi-modal features; and determine a processing result based on the target feature, the processing result being a processing result of the first target video and the first viewing data; A memory is configured to store programs required by the processor to perform the above processing process.

Images